Artificial intelligence-based teaching quality real-time evaluation system and method

Abstract

The invention relates to the technical field of education, in particular to a teaching quality real-time evaluation system and method based on artificial intelligence. Comprising a real-time portrait construction module used for obtaining a shooting picture of a classroom camera, carrying out real-time analysis on the shooting picture based on artificial intelligence, and constructing a real-time figure image; the teaching behavior recognition module is used for carrying out behavior recognition on the real-time figure image and determining a current teaching behavior; and the quality evaluation module is used for evaluating the teaching quality of the current teacher based on all teaching behaviors and a preset teaching quality evaluation model, so that teaching evaluation can be carried out on the teacher more objectively and accurately.

Description

Technical Field

[0001] The present invention relates to the field of educational technology, and in particular to a real-time teaching quality evaluation system and method based on artificial intelligence. Background Art

[0002] In today's education landscape, classroom instruction remains the core method for achieving effective teaching and learning. Accurate and scientific classroom teaching evaluation plays a key role in promoting student growth, facilitating teacher professional development, and improving teaching quality. Objectively and comprehensively evaluating teachers' teaching quality has become an integral part of classroom instruction.

[0003] Currently, common methods for evaluating classroom quality suffer from two significant problems. One relies on post-class student evaluations. While this method can reflect students' experience of the lecture, it is often significantly influenced by students' subjective preferences for the instructor. Due to personal preferences, students may give higher ratings to instructors they like and lower ratings to instructors they dislike, making the evaluation results less objective and fair and failing to truly reflect the instructor's teaching level. The other method relies on machine-generated quantitative scoring using video and audio recordings of classroom instruction. However, existing machine-generated scoring techniques typically only consider a limited number of factors: student listening status, engagement, and redundancy in lecture content. However, actual classroom situations are extremely complex and varied. These limitations make it difficult for evaluation results to accurately reflect the true quality of the class. The evaluation methods are also less robust and fail to fully capture key information from the teaching process.

[0004] The rapid development of artificial intelligence (AI), particularly the maturing of machine learning, data analysis, and computer vision, has created new opportunities for addressing the challenges of teaching quality assessment. Using AI to monitor and analyze the teaching process in real time can yield more comprehensive and accurate data, enabling a more objective assessment of teaching quality. Against this backdrop, a series of AI-based technologies have emerged, including the real-time portrait construction module, the teaching behavior recognition module, and the quality assessment module. By capturing images from classroom cameras, the real-time portrait construction module constructs real-time person images, and then uses the teaching behavior recognition module to identify teaching behaviors, the quality assessment module ultimately completes the assessment of the teacher's teaching quality. This system is expected to break through the limitations of traditional assessment methods and provide a more scientific and accurate solution for teaching quality assessment.

[0005] Therefore, the present invention provides a real-time teaching quality evaluation system and method based on artificial intelligence. Summary of the Invention

[0006] The present invention provides a real-time teaching quality evaluation system and method based on artificial intelligence, which is used to analyze the captured images through artificial intelligence, construct real-time character images, perform behavior recognition on the real-time character images, determine the current teaching behavior, and obtain the teacher's teaching quality evaluation based on all teaching behaviors and a preset teaching quality evaluation model, so as to evaluate the teacher's teaching more objectively and accurately.

[0007] The present invention provides a real-time teaching quality evaluation system based on artificial intelligence, comprising: A real-time portrait construction module acquires images captured by classroom cameras and performs real-time analysis of the images based on artificial intelligence to construct real-time portrait images; wherein the classroom cameras include teacher cameras and student cameras, and the real-time portrait images include real-time teacher portrait images and real-time student portrait images; A teaching behavior recognition module performs behavior recognition on the real-time character image to determine the current teaching behavior; wherein the teaching behavior includes: teacher behavior and student behavior; The quality assessment module evaluates the teaching quality of the current teacher based on all the teaching behaviors and the preset teaching quality assessment model.

[0008] Preferably, the real-time portrait construction module includes: a teacher shooting unit, which obtains a picture of the current teacher shot by the teacher camera, determines a first human multi-angle image and marks the current teacher; A human body tracing unit, performing human body contour tracing on the first multi-angle human body image to determine a human body contour map; a body part marking unit, for marking body parts on the human body outline image; The skeleton construction unit performs skeleton construction based on the body parts and a preset human skeleton library to determine the real-time teacher character image.

[0009] Preferably, the bone building unit comprises: A coordinate extraction unit, which obtains key points of the body part and extracts the two-dimensional coordinates of the key points; A coordinate conversion unit, converting the two-dimensional coordinates into normalized plane coordinates without distortion; A projection matrix construction unit, which obtains camera extrinsic parameters and constructs a projection matrix based on the extrinsic parameters; a three-dimensional coordinate determining unit, which calculates the three-dimensional coordinates according to the normalized plane coordinates and the projection matrix; A matching unit, which matches the three-dimensional coordinates with a preset human skeleton library; A dynamic skeleton model determination unit adjusts the skeleton parameters based on the matching results to generate a dynamic skeleton model; The completion subunit is used to retrieve the body parameters of the current teacher if the dynamic skeleton model is incomplete, complete the dynamic skeleton model, and obtain the real-time teacher character image.

[0010] Preferably, the real-time portrait construction module includes: a facial recognition unit, performing facial recognition on a current student according to the student camera to determine facial features of the current student; A facial matching unit, matching the facial features with a preset feature library; The student image acquisition unit acquires a second multi-angle image of the current student if the matching is unsuccessful; The student character image determining unit determines the real-time student character image based on the second multi-angle human body image.

[0011] Preferably, the teaching behavior recognition module includes: a posture analysis unit, performing a first human posture analysis on the real-time teacher character image to determine a first posture of the current teacher, and simultaneously performing a second human posture analysis on the real-time student character image to determine a second posture of the current student; The teaching behavior determining unit determines a current teaching behavior according to the first posture, the second posture and a preset posture-behavior table.

[0012] Preferably, the teaching behavior determination unit further includes: a behavior determination subunit, which determines a first teacher behavior and / or a first associated behavior based on the first posture and a preset teacher posture-behavior table, and simultaneously determines a first student behavior and / or a second associated behavior based on the second posture and a preset student posture-behavior table; The behavior association subunit is used to determine a second teacher behavior and a second student behavior according to the first associated behavior and the second associated behavior.

[0013] Preferably, the quality assessment module includes: a teacher behavior statistics unit, which collects statistics on all teacher behaviors based on preset teacher behavior types, and records the first positive behavior and the number of first positive behaviors, the first negative behavior and the number of first negative behaviors of the current teacher; a student behavior statistics unit, which collects statistics on all student behaviors based on preset student behavior types, and records the second positive behavior and the number of the second positive behaviors, the second negative behavior and the number of the second negative behaviors of the current student; The evaluation unit inputs the first positive behavior and the number of first positive behaviors, the first negative behavior and the number of first negative behaviors, the second positive behavior and the number of second positive behaviors, and the second negative behavior and the number of second negative behaviors into the teaching quality evaluation model to obtain the teaching quality evaluation of the current teacher.

[0014] Preferably, the padding subunit includes: a coarse comparison block, which performs a coarse comparison between the dynamic skeleton model and the human skeleton model to obtain a first missing bone; A coarse filling block generates the first missing bones according to the body parameters of the current teacher and performs coarse filling; a connection position determination block, locking a plurality of connection positions between the first missing skeleton and the dynamic skeleton model; The connected edge segmentation blocks are divided into a plurality of three-dimensional unit blocks by segmenting the connected edge area of ​​the dynamic skeleton model according to the three-dimensional unit size based on the corresponding connected positions; A surface curvature pair acquisition block is used to sequentially acquire the surface curvature pairs of the outer surface of each three-dimensional unit block, wherein the surface curvature set is obtained by capturing the curvature of the central surface line of the outer surface according to the priority principle from top to bottom and from left to right; An array determination block constructs a first array based on all surface radians involved in the connected edge region, and determines a second array of the connected region based on missing body parameters of the current teacher and the corresponding connected edge region that need to be filled in with bones;

[0015]

[0016] in, 、 They respectively represent the radian curvature of the central curved surface line from top to bottom and from left to right of the unit block connected to the i-th three-dimensional unit block in the adjacent edge area; The curvature of the arc of the center surface line from top to bottom of the unit block connecting the i-th three-dimensional unit block in the adjacent edge area; represents the arc curvature of the central curved surface line from top to bottom of the unit block after the missing unit blocks of the i-th three-dimensional unit block in the adjacent edge area are filled based on the body parameters of the current teacher; represents the regulation function from top to bottom; The curvature of the central curved line from left to right of the unit block connecting the i-th three-dimensional unit block in the adjacent edge area; represents the radian curvature of the central curved surface line from left to right of the unit block after the missing unit blocks of the i-th three-dimensional unit block in the adjacent edge area are filled based on the body parameters of the current teacher; represents the regulation function from left to right; r1 represents the curvature threshold; Indicates from Randomly select 1 value from the . Indicates from Randomly select 1 value from the . The padding optimization block performs padding optimization on the connection area according to the second array to obtain a padded skeleton model.

[0017] The present invention provides a real-time teaching quality evaluation method based on artificial intelligence, comprising: S101: Based on images captured by classroom cameras, and using artificial intelligence to perform real-time analysis on the captured images, constructing real-time character images; wherein the classroom cameras include a teacher camera and a student camera, and the real-time character images include a real-time character image of the teacher and a real-time character image of the students; S102: performing behavior recognition on the real-time person image to determine current teaching behavior; wherein the teaching behavior includes: teacher behavior and student behavior; S013: Based on all the teaching behaviors and the preset teaching quality evaluation model, conduct a teaching quality evaluation on the current teacher.

[0018] Compared with the prior art, the present invention has the following advantages: Through artificial intelligence, the captured images are analyzed, real-time character images are constructed, and behavior recognition is performed on the real-time character images to determine the current teaching behavior. Based on all teaching behaviors and the preset teaching quality evaluation model, the teacher's teaching quality evaluation is obtained, which can evaluate the teacher's teaching more objectively and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 4 is a structural diagram of a real-time teaching quality evaluation system based on artificial intelligence in an embodiment of the present invention; Figure 2 The figure is a flow chart of a method for real-time evaluation of teaching quality based on artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] Example 1: The embodiment of the present invention provides a real-time teaching quality evaluation system based on artificial intelligence, such as Figure 1 As shown, including: The real-time portrait construction module obtains the images captured by the classroom cameras and performs real-time analysis of the images based on artificial intelligence to construct real-time character images. The classroom cameras include the teacher camera and the student cameras, and the real-time character images include the real-time teacher image and the real-time student image. The teaching behavior recognition module performs behavior recognition on real-time character images to determine the current teaching behavior; teaching behavior includes: teacher behavior and student behavior; The quality assessment module evaluates the teaching quality of current teachers based on all teaching behaviors and the preset teaching quality assessment model.

[0023] In this embodiment, the real-time character image is a dynamic skeleton model of the character. The teacher camera and the student camera are arranged in advance based on the positions of the teacher and the students, and there are at least two cameras of each type.

[0024] In this example, teacher behaviors include walking up to the podium, asking questions, walking around, writing on the blackboard, and making phone calls. Student behaviors include listening, standing, playing with their phones, and lying on their desks. Teaching behaviors are generated by constructing dynamic skeletal models of the teacher or student, determining their posture, and then generating the corresponding teaching behaviors. For example, the teacher's dynamic skeletal model shows the teacher facing away from the students, raising their right hand and sliding it, resulting in the teacher writing on the blackboard behavior.

[0025] In this embodiment, all teaching behaviors are all teacher behaviors and student behaviors in a class. The preset teaching quality evaluation model is pre-trained based on assigning corresponding scores to various teaching behaviors, and inputs teaching behaviors and corresponding times, and outputs scores.

[0026] The beneficial effects of the above technical solution are: through artificial intelligence, the captured images are analyzed, real-time character images are constructed, behavior recognition is performed on the real-time character images, the current teaching behavior is determined, and based on all teaching behaviors and the preset teaching quality evaluation model, the teacher's teaching quality evaluation is obtained, which can evaluate the teacher's teaching more objectively and accurately.

[0027] Example 2: Based on Example 1, the real-time portrait construction module includes: The teacher shooting unit obtains the shooting picture of the current teacher by the teacher camera, determines the multi-angle image of the first human body and marks the current teacher; A human body tracing unit, performing human body contour tracing on the first human body multi-angle image to determine a human body contour map; Body part marking unit, marking body parts on the human body outline; The skeleton construction unit constructs skeletons based on body parts and a preset human skeleton library to determine the real-time teacher character image.

[0028] In this embodiment, the first human multi-angle image is a collection of teacher images taken from different perspectives by multiple teacher cameras, and the current teacher is marked for subsequent tracking and shooting.

[0029] In this embodiment, the scene and human body of each image in the teacher image set are segmented based on the deep learning model, the human body contour is extracted using the Canny algorithm to obtain a single human body contour map, and the contour maps of multiple perspectives are fused into a consistent three-dimensional contour map according to the three-dimensional contour reconstruction method.

[0030] In this embodiment, the body parts of the human body outline are marked with basic parts, including head, torso, hands, etc.

[0031] In this embodiment, the real-time teacher character image is obtained by obtaining the 2D coordinates of the joints of the body parts, performing distortion-free processing on the 2D coordinates to determine the normalized planar coordinates, constructing a projection matrix based on the camera's extrinsic parameters, calculating the 3D coordinates of the joints, and matching these 3D coordinates with a preset human skeleton library to obtain a dynamic skeleton model. The preset human skeleton library is a parametric human body model.

[0032] The beneficial effects of the above technical solution are: through the teacher's shooting picture, the multi-angle image of the human body is determined, the human body contour is outlined in the multi-angle image of the human body, the human body contour map is determined, the body parts of the human body contour map are marked, and based on the preset human skeleton library, the real-time teacher character image is determined, which lays the foundation for the subsequent determination of the teacher's behavior.

[0033] Example 3: Based on Example 2, the skeleton building unit includes: A coordinate extraction unit, which obtains key points of body parts and extracts the two-dimensional coordinates of the key points; A coordinate conversion unit converts two-dimensional coordinates into normalized plane coordinates without distortion; A projection matrix construction unit obtains camera extrinsic parameters and constructs a projection matrix based on the extrinsic parameters; A three-dimensional coordinate determination unit calculates the three-dimensional coordinates according to the normalized plane coordinates and the projection matrix; A matching unit, which matches the three-dimensional coordinates with the preset human skeleton library; A dynamic skeleton model determination unit adjusts the skeleton parameters based on the matching results to generate a dynamic skeleton model; The completion subunit is used to retrieve the body parameters of the current teacher if the dynamic skeleton model is incomplete, complete the dynamic skeleton model, and obtain a real-time teacher character image.

[0034] In this embodiment, the key points are joint points of body parts, which are obtained through a key point detection model set in advance, and the two-dimensional coordinates are the pixel coordinates of the key points.

[0035] In this embodiment, each camera is calibrated with internal parameters in advance using a checkerboard calibration plate. The internal parameter matrix K and the distortion coefficient dist are calculated by Zhang Zhengyou's calibration method. Based on the built-in matrix and distortion parameters, the two-dimensional pixel coordinates of the key points are dedistorted and converted into normalized plane coordinates.

[0036] In this embodiment, camera extrinsic parameters include a rotation matrix and a translation vector. Both camera intrinsic and extrinsic parameters are obtained based on the world coordinate system. P1 = K1 * [R1 | t1], where P1 represents the projection matrix of camera 1, K1 represents the adjustment coefficient, with a value range of (0, 1), R1 represents the rotation matrix of camera 1, and t1 represents the translation vector of camera 1.

[0037] In this embodiment, the normalized plane coordinates of the key points are converted into three-dimensional coordinates through a multi-view geometry triangulation algorithm.

[0038] In this embodiment, the preset human skeleton library is a parametric human body model, and its skeleton parameters include posture parameters and shape parameters. The detected key points are aligned with the joint points of the parametric human body model based on three-dimensional coordinates, and the posture parameters are optimized by inverse kinematics to minimize the key point projection error. Based on the human morphology statistical database, the shape parameters are constrained to be within a physiologically reasonable range. The inverse kinematics optimization adopts the Levenberg-Marquardt algorithm and adds joint motion angle restrictions, including: hip abduction angle range of 0°~45°, knee flexion angle range of 0°~135°; spinal torsion angle range of -30°~30°, shoulder joint external rotation angle range of 0°~90°.

[0039] In this embodiment, the skeleton parameters of the current teacher are adjusted by matching the posture parameters and shape parameters to determine the skeleton model of each body part, thereby obtaining a dynamic skeleton model of the whole body.

[0040] In this example, the dynamic skeletal model is incomplete due to obstructions such as tables and chairs in the classroom. Body parameters are parameters such as the height and weight of each teacher. Because the dynamic skeletal model may be incomplete due to obstructions such as tables and chairs in the classroom, the body parameters of the current teacher are used to complete the remaining skeletal morphology based on the partial dynamic skeletal model and body parameters.

[0041] In this embodiment, the real-time character image is a 3D image of a dynamic skeleton model in a classroom.

[0042] The beneficial effects of the above technical solution are: by converting the two-dimensional coordinates of the key points of the body parts, normalized plane coordinates are obtained, and based on the external parameter projection matrix, the three-dimensional coordinates of the key points are calculated, and matching is performed based on the three-dimensional coordinates and the preset human skeleton library, and the skeleton parameters are adjusted according to the matching results to generate a dynamic skeleton model. For the incomplete dynamic skeleton model, the teacher's body parameters are retrieved to complete it, and a real-time teacher character image is obtained. The teacher's behavior can be monitored in real time, laying the foundation for the subsequent determination of the teacher's behavior.

[0043] Example 4: Based on Example 1, the real-time portrait construction module includes: The facial recognition unit performs facial recognition on the current student based on the student camera to determine the facial features of the current student; A facial matching unit that matches facial features with a preset feature library; The student image acquisition unit acquires a second multi-angle image of the current student if the matching is unsuccessful; The student character image determining unit determines a real-time student character image based on the second multi-angle human body image.

[0044] In this embodiment, facial features include head posture and attention judgment.

[0045] In this embodiment, the pre-set feature library contains mappings between student behaviors and facial features. These behaviors include lying on the desk, turning around, playing with a phone, and other behaviors that can be directly identified by facial features. If a match is successful, a real-time student image is generated based on the behavior.

[0046] In this embodiment, the second multi-angle image of the human body is similar to the first multi-angle image of the human body. The determination of the real-time student character image is similar to the determination of the real-time teacher character image.

[0047] The beneficial effects of the above technical solution are: by matching the student's facial features with the preset feature library, if the match is successful, the real-time student image is directly determined; if the match is unsuccessful, the real-time image is determined based on the multi-angle image of the human body, which reduces the difficulty of monitoring the student's classroom situation and lays the foundation for the subsequent determination of student behavior.

[0048] Example 5: Based on Example 4, the teaching behavior recognition module includes: a posture analysis unit, performing a first human posture analysis on the real-time teacher character image to determine a first posture of the current teacher, and simultaneously performing a second human posture analysis on the real-time student character image to determine a second posture of the current student; The teaching behavior determination unit determines the current teaching behavior according to the first posture, the second posture and the preset posture-behavior table.

[0049] In this embodiment, the first posture is the posture of the teacher, including: sitting on a chair, standing facing the students, facing away from the students and raising hands, etc.

[0050] In this embodiment, the second posture is the student's posture, including bowing the head, standing, etc. The preset posture-behavior table includes a preset teacher posture-behavior table and a preset student posture-behavior table. Both are derived by artificial intelligence based on the specific behaviors of teachers and students in class to infer the postures that will be performed. The corresponding behavior can be obtained based on the postures.

[0051] The beneficial effects of the above technical solution are: by performing posture analysis on real-time character images, the postures of teachers and students are determined, and based on the preset posture-behavior table, the teaching behavior is determined, laying the foundation for the subsequent evaluation of the teacher's teaching quality.

[0052] Example 6: Based on Example 5, the teaching behavior determination unit further includes: a behavior determination subunit, which determines a first teacher behavior and / or a first associated behavior based on the first posture and a preset teacher posture-behavior table, and simultaneously determines a first student behavior and / or a second associated behavior based on the second posture and a preset student posture-behavior table; The behavior association subunit is used to determine a second teacher behavior and a second student behavior based on the first associated behavior and the second associated behavior.

[0053] In this embodiment, the preset teacher posture-behavior table is pre-set by the AI ​​based on the teacher's behavior during class and the postures corresponding to those behaviors. First teacher behaviors are behaviors where the teacher does not interact with students, such as writing on the blackboard or turning their back to students. These behaviors can be directly obtained by matching the teacher's posture with the preset teacher posture-behavior table. First associated behaviors are behaviors where the teacher interacts with students, such as answering questions or asking questions.

[0054] In this embodiment, the preset student posture-behavior table is pre-set by the AI ​​based on students' behavior during class and their corresponding postures. The first student behavior is non-interactive behavior between the student and the teacher, such as playing with a phone or lying on the desk. This can be directly obtained by matching the student's posture with the preset student posture-behavior table. The second associated behavior is interactive behavior between the student and the teacher, such as answering questions or asking questions.

[0055] In this embodiment, the first associated behavior derived from the teacher's posture does not necessarily indicate interaction between the teacher and the student. Therefore, it is necessary to determine whether the student has a corresponding second associated behavior to ensure that the teacher and the student are interacting. For example, if the first associated behavior is the teacher asking a question, but there is no corresponding second associated behavior of the student standing up, then the first associated behavior is not a teacher behavior.

[0056] The beneficial effects of the above technical solution are: determining teacher behavior and teacher-related behavior through teacher posture and preset teacher posture-behavior table, determining student behavior and student-related behavior according to student posture and preset student posture-behavior table, and determining whether the associated behavior is teacher behavior and student behavior based on teacher-related behavior and student-related behavior, so as to more accurately obtain the behavior of teachers and students in the classroom.

[0057] Example 7: Based on embodiment 2 or 4, the quality assessment module includes: A teacher behavior statistics unit collects statistics on all teacher behaviors based on preset teacher behavior types, and records the first positive behavior and the number of first positive behaviors, the first negative behavior and the number of first negative behaviors of the current teacher; A student behavior statistics unit collects statistics on all student behaviors based on preset student behavior types, and records the second positive behavior and the number of second positive behaviors, the second negative behavior and the number of second negative behaviors of the current student; The evaluation unit inputs the first positive behavior and the number of the first positive behavior, the first negative behavior and the number of the first negative behavior, the second positive behavior and the number of the second positive behavior, and the second negative behavior and the number of the second negative behavior into the teaching quality evaluation model to obtain the teaching quality evaluation of the current teacher.

[0058] In this embodiment, the preset teacher behavior types include: answering or making phone calls, facing students, turning away from students, writing on the blackboard, walking down the podium, etc. The first positive behavior is consistent with the behavior of a teacher during class, including: facing students, walking down the podium, etc. The first negative behavior is inconsistent with the behavior of a teacher during class, including: answering or making phone calls, smoking, etc.

[0059] In this embodiment, the preset student behavior types include: lying on the table, playing with mobile phones, reading and writing, listening to lectures, etc. The second positive behavior is consistent with the behavior of students in class, including: listening to lectures, reading and writing, raising hands, etc. The second negative behavior is inconsistent with the behavior of students in class, including: lying on the table, playing with mobile phones, etc.

[0060] In this embodiment, the teaching quality evaluation is reflected in the form of a score, which can be a percentage system or a ten-point system.

[0061] The beneficial effects of the above technical solution are: by pre-setting teacher behavior types, teacher behaviors are divided into positive behaviors and negative behaviors, and by pre-setting student behavior models, student behaviors are divided into positive behaviors and negative behaviors. Positive behaviors and the number of positive behaviors, negative behaviors and the number of negative behaviors are input into the teaching quality evaluation model to obtain the teacher's teaching quality evaluation, and the teacher's teaching quality evaluation can be conducted more objectively and accurately based on the teacher's usual classroom performance.

[0062] Example 8: Based on Example 3, the subunits are completed, including: A coarse comparison block performs a coarse comparison between the dynamic skeleton model and the human skeleton model to obtain the first missing bone; The coarse filling block generates the first missing bone according to the current teacher's body parameters and performs coarse filling; A connection position determination block is used to lock a plurality of connection positions between the first missing bone and the dynamic bone model; The connected edge segmentation block is based on the corresponding connected position, and the connected edge area of ​​the dynamic skeleton model is segmented according to the three-dimensional unit size to obtain a number of three-dimensional unit blocks; A surface curvature pair acquisition block is used to sequentially acquire the surface curvature pairs of the outer surface of each three-dimensional unit block. The surface curvature pairs are obtained by capturing the curvature of the central surface line of the outer surface in a top-to-bottom and left-to-right priority manner. The array determination block constructs a first array based on all surface curvatures involved in the connected edge area, and determines a second array of the connected area based on missing body parameters of the current teacher and the corresponding connected edge area that need to be filled in with bones;

[0063]

[0064] in, 、 They respectively represent the radian curvature of the central curved surface line from top to bottom and from left to right of the unit block connected to the i-th three-dimensional unit block in the adjacent edge area; The curvature of the arc of the center surface line from top to bottom of the unit block connecting the i-th three-dimensional unit block in the adjacent edge area; represents the arc curvature of the central surface line from top to bottom of the unit block after the missing unit blocks of the i-th three-dimensional unit block in the adjacent edge area are filled based on the body parameters of the current teacher; represents the regulation function from top to bottom; The curvature of the central curved line from left to right of the unit block connecting the i-th three-dimensional unit block in the adjacent edge area; represents the arc curvature of the central surface line from left to right of the unit block after the missing unit blocks of the i-th three-dimensional unit block in the adjacent edge area are filled based on the body parameters of the current teacher; represents the regulation function from left to right; r1 represents the curvature threshold; Indicates from Randomly select 1 value from the . Indicates from Randomly select 1 value from the . The padding optimization block performs padding optimization on the connection area according to the second array to obtain a padded skeleton model.

[0065] In this embodiment, the first missing bone is the missing portion of the dynamic skeletal model obtained by aligning and comparing the dynamic skeletal model with the human skeletal model. For example, if the dynamic skeletal model represents the teacher's upper body, after a rough comparison, the first missing bone is the teacher's lower body. Alternatively, assuming a standard human skeletal model has 206 bones, the dynamic skeletal model captures 200 bones in real time. During the comparison process, it is discovered that several small bones at the wrist (such as the pisiform bones) are missing from the dynamic skeletal model. After a traversal search, these bones are determined to be the first missing bones. Assuming the current teacher's height is 175 cm, based on an existing database of height-bone length ratios (for example, for a 175 cm tall person, the pisiform bones at the wrist are approximately 1.2 cm long), a cylinder generation function in the 3D modeling software (because the pisiform bones are approximately cylindrical) is used, with a radius of 0.6 cm and a height of 1.2 cm. This creates a preliminary 3D model of the pisiform bones, completing the rough completion of this missing bone.

[0066] Taking the wrist bone connection as an example, the missing pisiform bone in the dynamic skeletal model is connected to the triquetrum. Through spatial geometric analysis, the positional relationship between the triquetrum and pisiform bones in three-dimensional space is calculated. Assuming the triquetrum's spatial coordinates are (10, 10, 10), the pisiform bone's connection location is at (10, 10, 11) on the triquetrum (unit: centimeters). Through calculation and logical judgment, this location is determined to be a connected connection. In this embodiment, the connected edge region is the region of the dynamic skeletal model within the connected connection location. The 3D unit size is pre-set, and the 3D unit block is the 3D region of the dynamic skeletal model's connected edge. Assuming the 3D unit size is 1cm×1cm×1cm, the edge region where the triquetrum and pisiform bones meet is segmented. This connected edge region is a 3cm×3cm×3cm cube. According to the algorithm, it is segmented into 27 1cm×1cm×1cm 3D unit blocks.

[0067] In this embodiment, the first array is constructed based on the position of all surface curvature pairs. The missing body parameters include the lengths of the bones of the missing part, and the connection area is the area of ​​the first missing bone in the connected position. The second array is constructed based on the position of the surface curvature pairs obtained by extending the first array to the connection area based on artificial intelligence. For one of the three-dimensional unit blocks, the center curve of its outer surface is assumed to be an arc. By sampling five points on the curve, the curvatures of these five points are calculated to be 0.2, 0.3, 0.4, 0.3, and 0.2 (unit: radians), and the curvatures are 0.5, 0.6, 0.7, 0.6, and 0.5 (unit: 1 / cm), respectively. The surface curvature pairs are formed in the order ((0.2, 0.5), (0.3, 0.6), (0.4, 0.7), (0.3, 0.6), and (0.2, 0.5)). Arrange the curvature pairs of the 27 three-dimensional unit blocks in the first array, from left to right, top to bottom, and front to back. Suppose, after analysis, it is found that the teacher's wrist bones are 0.5 cm shorter than the standard model. Based on this missing body parameter, a series of adjustment parameters are calculated to form the second array, which will be used for precise adjustment when the bones are subsequently filled in. The number of curvature pairs in the second array is the same as in the first array, and they are placed in a one-to-one correspondence.

[0068] In this embodiment, the posture of the first missing bone is adjusted based on the surface curvature of each three-dimensional unit block of the second array. When the surface curvature of the first missing bone of a certain posture in the connection area matches the second array, the first missing bone is adjusted to complete the bone model for the posture.

[0069] In this embodiment, during the skeleton completion process, due to the varying physical parameters of different teachers and the complex and diverse curved shapes of the connecting edge regions, randomly selecting the original curvature values ​​(d1i, u1i, d2i, u2i) and combining them with an adjustment function can simulate and adapt to actual skeletal surface variations to a certain extent, making the generated skeletal surface in the connecting region more closely aligned with the current teacher's physical characteristics and increasing the model's flexibility and accuracy. When processing the connecting edge regions, the relationship between the different original curvature values ​​(d1i and u1i) and the curvature threshold r1 reflects the varying degrees of curvature and characteristics of the surface in that region. By designing a case-by-case adjustment function, appropriate adjustments can be made to address different surface curvature conditions. This allows for effective adjustment of the surface curvature when constructing the connecting region, ensuring that the completed skeleton's surface morphology better meets actual requirements and improving the accuracy and consistency of the skeletal model. In the left-to-right direction, the relationship between the different original curvature values ​​(d2i and u2i) and the curvature threshold r1 reflects the varying characteristics of the surface in that direction. By designing adjustment functions according to different situations, the curvature of the surface in the connection area from left to right can be reasonably adjusted according to the actual curvature of the surface, so that the surface shape of the completed skeleton in this direction is more in line with reality, ensuring the accuracy and consistency of the skeleton model in all directions, and better adapting to the teacher's body parameters.

[0070] The beneficial effects of the above technical solution are: the missing bones in the dynamic skeleton model are filled in through the teacher's body parameters, and based on the connection position between the missing bones and the dynamic skeleton model, the surface curvature pairs of several three-dimensional unit blocks in the connected edge area of ​​the dynamic skeleton model are determined, and based on the body parameters, the surface curvature pairs of the connection area of ​​the missing bones are obtained, the missing bones are adjusted, and the dynamic skeleton model is filled in. It can effectively deal with the situation of body occlusion, construct a complete skeleton model, and accurately obtain the behavior of teachers and students in the classroom.

[0071] Example 9: A real-time teaching quality evaluation method based on artificial intelligence, such as Figure 2 As shown, including: S101: Based on images captured by classroom cameras, and using artificial intelligence, real-time analysis of the captured images is performed to construct real-time character images; wherein the classroom cameras include a teacher camera and a student camera, and the real-time character images include a real-time character image of the teacher and a real-time character image of the students; S102: Performing behavior recognition on the real-time person image to determine the current teaching behavior; wherein the teaching behavior includes: teacher behavior and student behavior; S103: Based on all teaching behaviors and the preset teaching quality evaluation model, the current teacher’s teaching quality is evaluated.

[0072] The beneficial effects of the above technical solution are: through artificial intelligence, the captured images are analyzed, real-time character images are constructed, behavior recognition is performed on the real-time character images, the current teaching behavior is determined, and based on all teaching behaviors and the preset teaching quality evaluation model, the teacher's teaching quality evaluation is obtained, which can evaluate the teacher's teaching more objectively and accurately.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A real-time teaching quality evaluation system based on artificial intelligence, characterized by: include: A real-time portrait construction module acquires images captured by classroom cameras and performs real-time analysis of the images based on artificial intelligence to construct real-time portrait images; wherein the classroom cameras include teacher cameras and student cameras, and the real-time portrait images include real-time teacher portrait images and real-time student portrait images; A teaching behavior recognition module performs behavior recognition on the real-time character image to determine the current teaching behavior; wherein the teaching behavior includes: teacher behavior and student behavior; The quality assessment module evaluates the teaching quality of the current teacher based on all the teaching behaviors and the preset teaching quality assessment model.

2. The real-time teaching quality evaluation system based on artificial intelligence according to claim 1 is characterized in that: The real-time portrait construction module includes: a teacher shooting unit, which obtains a picture of the current teacher shot by the teacher camera, determines a first human multi-angle image and marks the current teacher; A human body tracing unit, performing human body contour tracing on the first multi-angle human body image to determine a human body contour map; a body part marking unit, for marking body parts on the human body outline image; The skeleton construction unit performs skeleton construction based on the body parts and a preset human skeleton library to determine the real-time teacher character image.

3. The real-time teaching quality evaluation system based on artificial intelligence according to claim 2 is characterized in that: The skeleton building unit comprises: A coordinate extraction unit, which obtains key points of the body part and extracts the two-dimensional coordinates of the key points; A coordinate conversion unit, converting the two-dimensional coordinates into normalized plane coordinates without distortion; A projection matrix construction unit, which obtains camera extrinsic parameters and constructs a projection matrix based on the extrinsic parameters; a three-dimensional coordinate determining unit, which calculates the three-dimensional coordinates according to the normalized plane coordinates and the projection matrix; A matching unit, which matches the three-dimensional coordinates with a preset human skeleton library; A dynamic skeleton model determination unit adjusts the skeleton parameters based on the matching results to generate a dynamic skeleton model; The completion subunit is used to retrieve the body parameters of the current teacher if the dynamic skeleton model is incomplete, complete the dynamic skeleton model, and obtain the real-time teacher character image.

4. The real-time teaching quality evaluation system based on artificial intelligence according to claim 1 is characterized in that: The real-time portrait construction module includes: a facial recognition unit, performing facial recognition on a current student according to the student camera to determine facial features of the current student; A facial matching unit, matching the facial features with a preset feature library; The student image acquisition unit acquires a second multi-angle image of the current student if the matching is unsuccessful; The student character image determining unit determines the real-time student character image based on the second multi-angle human body image.

5. The real-time teaching quality evaluation system based on artificial intelligence according to claim 4 is characterized in that: The teaching behavior recognition module includes: a posture analysis unit, performing a first human posture analysis on the real-time teacher character image to determine a first posture of the current teacher, and simultaneously performing a second human posture analysis on the real-time student character image to determine a second posture of the current student; The teaching behavior determining unit determines a current teaching behavior according to the first posture, the second posture and a preset posture-behavior table.

6. The real-time teaching quality evaluation system based on artificial intelligence according to claim 5 is characterized in that: The teaching behavior determination unit further includes: a behavior determination subunit, which determines a first teacher behavior and / or a first associated behavior based on the first posture and a preset teacher posture-behavior table, and simultaneously determines a first student behavior and / or a second associated behavior based on the second posture and a preset student posture-behavior table; The behavior association subunit is used to determine a second teacher behavior and a second student behavior according to the first associated behavior and the second associated behavior.

7. The real-time teaching quality evaluation system based on artificial intelligence according to claim 4 is characterized in that: The quality assessment module includes: a teacher behavior statistics unit, which collects statistics on all teacher behaviors based on preset teacher behavior types, and records the first positive behavior and the number of first positive behaviors, the first negative behavior and the number of first negative behaviors of the current teacher; a student behavior statistics unit, which collects statistics on all student behaviors based on preset student behavior types, and records the second positive behavior and the number of the second positive behaviors, the second negative behavior and the number of the second negative behaviors of the current student; The evaluation unit inputs the first positive behavior and the number of first positive behaviors, the first negative behavior and the number of first negative behaviors, the second positive behavior and the number of second positive behaviors, and the second negative behavior and the number of second negative behaviors into the teaching quality evaluation model to obtain the teaching quality evaluation of the current teacher.

8. The real-time teaching quality evaluation system based on artificial intelligence according to claim 3 is characterized in that: The padding subunit includes: a coarse comparison block, which performs a coarse comparison between the dynamic skeleton model and the human skeleton model to obtain a first missing bone; A coarse filling block generates the first missing bones according to the body parameters of the current teacher and performs coarse filling; a connection position determination block, locking a plurality of connection positions between the first missing skeleton and the dynamic skeleton model; The connected edge segmentation blocks are divided into a plurality of three-dimensional unit blocks by segmenting the connected edge area of ​​the dynamic skeleton model according to the three-dimensional unit size based on the corresponding connected positions; A surface curvature pair acquisition block is used to sequentially acquire the surface curvature pairs of the outer surface of each three-dimensional unit block, wherein the surface curvature pairs are obtained by capturing the curvature of the central curved surface line of the outer surface according to the priority principle from top to bottom and from left to right; An array determination block constructs a first array based on all surface radians involved in the connected edge region, and determines a second array of the connected region based on missing body parameters of the current teacher and the corresponding connected edge region that need to be filled in with bones; in, 、 They respectively represent the radian curvature of the central curved surface line from top to bottom and from left to right of the unit block connected to the i-th three-dimensional unit block in the adjacent edge area; The curvature of the arc of the center surface line from top to bottom of the unit block connecting the i-th three-dimensional unit block in the adjacent edge area; represents the arc curvature of the central curved surface line from top to bottom of the unit block after the missing unit blocks of the i-th three-dimensional unit block in the adjacent edge area are filled based on the body parameters of the current teacher; represents the regulation function from top to bottom; The curvature of the central curved line from left to right of the unit block connecting the i-th three-dimensional unit block in the adjacent edge area; represents the radian curvature of the central curved surface line from left to right of the unit block after the missing unit blocks of the i-th three-dimensional unit block in the adjacent edge area are filled based on the body parameters of the current teacher; represents the regulation function from left to right; r1 represents the curvature threshold; Indicates from Randomly select 1 value from the . Indicates from Randomly select 1 value from the . The padding optimization block performs padding optimization on the connection area according to the second array to obtain a padded skeleton model.

9. A real-time teaching quality evaluation method based on artificial intelligence, characterized in that: include: S101: Based on images captured by classroom cameras, and using artificial intelligence to perform real-time analysis on the captured images, constructing real-time character images; wherein the classroom cameras include a teacher camera and a student camera, and the real-time character images include a real-time character image of the teacher and a real-time character image of the students; S102: performing behavior recognition on the real-time person image to determine current teaching behavior; wherein the teaching behavior includes: teacher behavior and student behavior; S103: Based on all the teaching behaviors and the preset teaching quality evaluation model, the teaching quality of the current teacher is evaluated.

Images